Snorkel shows partial credit in RL can oppose completion, proposes binary-gated GRPO
ajratner · x · 2026-10-16
Snorkel AI researchers show that partial-credit rewards in agent RL can work against the goal they serve: a policy that always makes partial progress but never finishes (pass@k = 0) can score a higher fractional return (0.50) than one that completes 20% of the time. Naive partial credit teaches models to satisfy easy checks while abandoning the path that satisfies them all, and blending it with pass/fail doesn't fix this unless nearly all weight sits on pass/fail. They introduce binary-gated GRPO, which removes the conflict while keeping partial credit's benefits; with it, harm requires a specific combination of conditions rare in natural settings.
More from Research
- RFD4-Proteina's 'underrated' docs detail partial diffusion for protein redesign — AllThingsApx · 2026-10-23
- MPI-Sec PhD Thesis on Vulnerability Detection Benchmarks Includes Pre-LLM Classic "Top Score on the Wrong Exam" — mboehme_ · 2026-10-23
- LiveNerf's 29-day data finds no nerfing of Claude Opus 5.5 — TheOnlyVibemaster · 2026-10-23
- New paper makes 9 falsifiable predictions toward a major AI-enabled terror attack by 2027-2028 — StephenLCasper · 2026-10-23
- HumanoidPose recovers humanoid robots' full-body joint states from RGB video alone — cholz · 2026-10-23
- Sophont's Holon-S1 predicts genes and biomarkers from routine pathology slides — iScienceLuvr · 2026-10-23